I have been wondering about the "Checking What the Engine Receives" issue for a while now, and am still unclear. I see tech writing articles tersely state how DITA/semantic tags help AI engines—it's easy to see the logic there. However, my understanding is that DITA tags are not present on the websites (aka CDPs) where published content lives (it all transforms to HTML). And, wouldn't most AI answer engines just be looking at the website content (without the DITA tags)?
Perhaps there is an easy answer that I'm missing? Are folks pushing raw DITA to their publishing site (similar to other behind-the-scenes metadata) and somehow wiring up the AI engine to look at that instead of the HTML content? If so, isn't this a bit of a major paradigm shift that CMS companies and tooling ppl should be making noise about?
I would like someone to break down in nuts and bolts terms the practical ways that you would have an AI engine on your docs website keying off DITA tags that live in your authoring tool.
That's awesome, thanks for the response. If it is true that DITA tagging is a major boon to making AI effective, it seems like a big opportunity for the CMS companies to provide an out-of-the-box authoring and publishing stack that easily ties together your tagging and an AI agent. Maybe they are working on it?
Such a solution could slow the move toward docs-as-code stacks, although there is another tech writing website that explains how you can somehow implement DITA tags via Oxygen alongside markdown files in a docs-as-code environment.
Your psychic power is strong ("maybe they ARE working on it") is a good guess. And, I'd wager lots of tech docs teams are also working on this using agentic AI. My psychic power tells me there will be several approaches presented to us as options, but there's only one viable one (the winner) and firms that can't risk turning their AI efforts into another giant content hairball will adopt.
This is both an interesting time filled with opportunity for big useful changes in our discipline and also a terrifyingly confusing time filled with opportunity for things to go terribly wrong.
I have been wondering about the "Checking What the Engine Receives" issue for a while now, and am still unclear. I see tech writing articles tersely state how DITA/semantic tags help AI engines—it's easy to see the logic there. However, my understanding is that DITA tags are not present on the websites (aka CDPs) where published content lives (it all transforms to HTML). And, wouldn't most AI answer engines just be looking at the website content (without the DITA tags)?
Perhaps there is an easy answer that I'm missing? Are folks pushing raw DITA to their publishing site (similar to other behind-the-scenes metadata) and somehow wiring up the AI engine to look at that instead of the HTML content? If so, isn't this a bit of a major paradigm shift that CMS companies and tooling ppl should be making noise about?
I would like someone to break down in nuts and bolts terms the practical ways that you would have an AI engine on your docs website keying off DITA tags that live in your authoring tool.
ShaneOhooligan:
I'm on it. Working on a post and some explainer videos to tackle this very subject. Thanks for bringing that up.
There are various approaches being used and lots of experimentation. More to come. Watch this space.
That's awesome, thanks for the response. If it is true that DITA tagging is a major boon to making AI effective, it seems like a big opportunity for the CMS companies to provide an out-of-the-box authoring and publishing stack that easily ties together your tagging and an AI agent. Maybe they are working on it?
Such a solution could slow the move toward docs-as-code stacks, although there is another tech writing website that explains how you can somehow implement DITA tags via Oxygen alongside markdown files in a docs-as-code environment.
Your psychic power is strong ("maybe they ARE working on it") is a good guess. And, I'd wager lots of tech docs teams are also working on this using agentic AI. My psychic power tells me there will be several approaches presented to us as options, but there's only one viable one (the winner) and firms that can't risk turning their AI efforts into another giant content hairball will adopt.
This is both an interesting time filled with opportunity for big useful changes in our discipline and also a terrifyingly confusing time filled with opportunity for things to go terribly wrong.
Either way, we've got work to do. :)